كشف ثغرات البرمجيات باستخدام التعلم العميق

Software Vulnerabilities Detection Using Deep Learning

الباحث: شهد جلال أحمد الطائي
المشرف: دجان بشير طه
الجامعة: جامعة الموصل
الكلية / القسم: كلية علوم الحاسوب والرياضيات / البرمجيات
الشهادة: ماجستير
السنة: 2023
لغة الرسالة: عربي
الكلمات المفتاحية: ثغرات البرمجيات ; التعلم العميق ; أمن البرمجيات ; الشبكات العصبية

Abstract

The process of detecting software vulnerabilities is considered a significant challenge for researchers, as they strive to achieve accuracy and excellent performance in this vital field. These challenges arise from the urgent need to discover vulnerabilities in a wide range of domains dealing with digital documents, including software security and protection against intrusions and other security risks. Additionally, technological advancements and the increasing complexity of software continue to make research in this field more difficult, necessitating urgent efforts and investment in research and innovation to keep up with advanced cyber threats. Continuous efforts and investment in research and innovation are required to develop advanced techniques that contribute to the accurate and effective discovery of software vulnerabilities. Researchers need the necessary skills to deal with software complexities and comprehensively test them to identify and fix potential vulnerabilities. In this study, a hybrid algorithm combining the advantages of Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) for reading and classifying sequential data in software functions. This algorithm excels in handling sequentially varying input lengths, allowing it to deal with different dimensions in software functions without the need to specify or indicate those dimensions during the training phase. The Draper database, containing 1.27 million functions extracted from open-source software, was used to train this algorithm. Providing a sufficient amount of data for training and using pre-processing techniques are crucial factors in building a strong and effective model in this field. The results demonstrated that the proposed model achieves significantly higher accuracy compared to previous researches, with an accuracy rate of 93.53%. The recall rate reached 90%, while the precision rate was 62.07%. The F1-Score, which combines precision and recall, was calculated for the model and reached 73.47% in the process of identifying software vulnerabilities.

طلب نسخة من الملف

الملفات غير منشورة على الموقع. اترك بياناتك وسيتواصل معك موظفو قسم الشؤون الثقافية والفنون لإرسال النسخة مجاناً.